Image retrieval based on entropy and invariant moments
Junding Sun · 2007
A new image retrieval algorithm based on entropy and invariant moments was presented. Though entropy was widely used in image retrieval, it didn't contain any spatial information. To solve such problem, an image was firstly partitioned into equal-sized sub-images. Then, unit entropy was introduced to describe its spatial feature based on the definition of entropy. Combined with unit entropy, an image was transformed to an entropy matrix. Then, Hu invariant moments and five new generalization invariant moments of the entropy matrix were calculated and normalized as an index. It is shown through the experiments that this algorithm is scale, rotation and translation invariant. In addition, the algorithm is not sensitive to image noise and the change of the image gray.